Which document workflow should you prepare first?

The instinct is to start with the worst one. The 400-page master service agreements, the vendor packet stuck between four teams since June, the folder everybody refers to by a nickname. Resist it. The first workflow you prepare should be the one that tells you something true about your own operation inside a couple of weeks, and the worst one rarely does. Three tests will get you to a better candidate.

Test one: is the reconciling the cost, or the reading?

Ask what people are actually doing with the time. If someone opens a document, reads it once, and acts, that process is already about as cheap as it gets. There is not much for a document platform to take off the table.

The processes worth preparing first look different. Someone opens the document, then opens a second document to check it against, then a spreadsheet that does not say which clause moved the number, then a third system for the receipt. The reading took four minutes. The rest of the afternoon went to reconciling, and nobody has ever written that time down as a line item because it does not feel like a task. It feels like the job.

A useful signal: how often does the answer live in one place? If a question about a renewal date sends someone into a lease, two amendments, and a rent roll, that is a reconciliation problem, and it is precisely what a prepared record removes. If the answer is on page one of one file, you are automating a lookup, and you will not feel the difference.

Test two: does the decision at the end have a named owner?

Every document workflow terminates in somebody deciding something. Payment gets approved. A discrepancy gets called material or immaterial. A vendor gets onboarded. A pay application moves or waits. If you cannot name the person who makes that call today — an actual person, not a committee or a step in a diagram — the workflow is not ready to be prepared, because there is nobody to hand the shorter list to.

This test does more work than it looks like it does. Alembic prepares the record and keeps unresolved work visible; the consequential decisions stay with the people your organization has authorized to make them. That arrangement only pays off when those people exist and are known. A workflow with a clear owner turns a queue into their queue. A workflow with a diffuse owner turns a queue into a nicer-looking version of the same standoff.

It also gives you your success measure for free. Ask the owner what they want on their desk in the morning. Usually the answer is not "all of it" — it is the exceptions, with the evidence attached, and some indication of what is still missing. That is a specification, and it came from the person who has to live with the result.

Test three: can you still count the exceptions?

Start where the failure modes are enumerable. In a good first workflow, someone can sit down and tell you the five or six ways a packet goes wrong: the arithmetic does not tie, the quantity disagrees with the receiving report, the certificate has expired, the sign-off is blank, the supporting document never arrived. Those are configurable checks, and each one you name is an exception the reviewer stops having to find by hand.

If nobody can list the failure modes, you have picked a process that is not understood yet. That is worth fixing, but it is a different project and it should not be the thing you are learning a new platform on.

Bounded volume helps too. A month of one document type in one team beats a year of everything. You want enough packets that the pattern is real and few enough that you can look at all of them when something surprises you.

Before you change it, measure the version you have

Take a bounded sample — twenty invoices, ten pay applications, one property's packet — and write down what actually happens. Minutes genuinely spent preparing each one, how often context was missing, how many correction loops it took, how long an exception sat before someone touched it. Do it on your own documents, with your own people, before anything changes.

This is unglamorous and it is the only honest baseline you will get. Industry averages are somebody else's operation. A vendor's savings percentage is a number chosen to be persuasive. Your own twenty-packet count is neither, and it is the thing you will point at in six months when somebody asks whether this was worth doing.

Where to look

Fields, tables, document types, terminology, checks, and outcome labels are configured around what you receive. Alembic can be configured to extract the fields and tables your review requires, rather than shipping a fixed opinion about what a lease or an invoice contains. So the question is genuinely which process first, not which of our products fits you.

Four common starting points are laid out in detail, each with its documents, its exceptions, and the decisions that stay with your people. Accounts payable is the one most teams can start on their own, because the volume is high and the exceptions are countable — that is the accounting and AP workflow. Lease and rent-roll reconciliation is the strongest case where the reconciling dwarfs the reading, on the commercial real estate page. Pay-application packets are the clearest example of a process gated by its least-ready attachment, on the construction page. And a vendor packet crossing legal, security, procurement, and finance is on the SaaS contracts page.

If none of those is quite your process, that is common and not a problem. Start from the four workflows, find the one whose shape matches yours, and bring us the packet and the decision it feeds. If it does not fit, we will say so.